全面提示词构建器 — 完整框架

🌟 创作者 🌱 Anonymous · NeLocker 社区

把任何主题变成详细、结构化的提示词，包含角色、背景、分隔符、少样本示例和边界情况处理。

标题和描述为机器翻译。提示词本身保留英文原文——AI 模型能理解任何语言；加上 “Answer in 简体中文.” 即可用你的语言获得回复。原文（英文）： The Expansive Prompt Builder — Full Scaffold — Turns any topic into a detailed, structured prompt using role, context, delimiters, few-shot examples, and edge-case handling.

更新于 2026年7月2日

为什么有效： It front-loads role and context to set expertise and voice, uses XML delimiters to separate instructions from input so the model never confuses them, and pairs positive rules with negative ones. Few-shot examples and explicit output formatting lock in structure, while edge-case handling prevents hallucination when information is missing.

💡 原版提示词——适用于任何模型。从这里开始，再针对各个模型微调。

# ROLE
You are a [specific expert role — e.g., "senior copywriter," not just "writer"] with deep expertise in [domain — the narrower the better, e.g., "B2B SaaS email marketing"]. You [key behavioral trait that shapes HOW it works — e.g., "prioritize clarity over cleverness," "always surface tradeoffs before recommending"].

# CONTEXT
[Background the model can't infer. The more real detail, the better the output.]
- Audience: [who reads/uses the output — e.g., "time-poor CTOs at Series A startups"]
- Purpose: [what success looks like — e.g., "book a 15-min demo call"]
- Constraints: [hard limits — budget, word count, tone, tech stack, brand voice, legal]

# TASK
[ONE directive, starts with an action verb — "Write," "Analyze," "Refactor," "Generate."]

Specifically:
1. [Concrete sub-step — break the task into checkable pieces]
2. [Sub-step]
3. [Sub-step]

# INPUT
[The raw material to act on. Keep it separate from instructions using the tags below.]
<input>
[Paste the actual content here — the article to summarize, code to review, data to analyze]
</input>

# REQUIREMENTS
- [Hard rule the output MUST follow — e.g., "cite every factual claim"]
- [Format/style rule — e.g., "active voice, no jargon"]
- [Scope boundary / negative rule — e.g., "do NOT invent statistics or quotes"]

# REASONING
Think step by step before answering. [Optional: "Show reasoning in <thinking> tags, then give the final answer in <answer> tags." — use for analytical or multi-step tasks]

# OUTPUT FORMAT
[Exactly how the answer should be structured — e.g., "JSON matching this schema," "3 markdown sections: Summary / Details / Next Steps," "under 150 words," "a table with columns X, Y, Z"]

# EXAMPLES
<example>
Input: [a short sample input]
Output: [the ideal response — teaches tone + structure faster than any description]
</example>

# EDGE CASES
- If [likely tricky condition], then [exact behavior you want].
- If information is missing, [ask a clarifying question / state assumptions / flag the gap].

已针对 Universal 调整格式

🔀 改编这条提示词——变成你自己的改编会注明出处，并让你的花园不断生长